rf_map.c revision 1.31 1 1.31 oster /* $NetBSD: rf_map.c,v 1.31 2004/03/07 02:59:25 oster Exp $ */
2 1.1 oster /*
3 1.1 oster * Copyright (c) 1995 Carnegie-Mellon University.
4 1.1 oster * All rights reserved.
5 1.1 oster *
6 1.1 oster * Author: Mark Holland
7 1.1 oster *
8 1.1 oster * Permission to use, copy, modify and distribute this software and
9 1.1 oster * its documentation is hereby granted, provided that both the copyright
10 1.1 oster * notice and this permission notice appear in all copies of the
11 1.1 oster * software, derivative works or modified versions, and any portions
12 1.1 oster * thereof, and that both notices appear in supporting documentation.
13 1.1 oster *
14 1.1 oster * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
15 1.1 oster * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
16 1.1 oster * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
17 1.1 oster *
18 1.1 oster * Carnegie Mellon requests users of this software to return to
19 1.1 oster *
20 1.1 oster * Software Distribution Coordinator or Software.Distribution (at) CS.CMU.EDU
21 1.1 oster * School of Computer Science
22 1.1 oster * Carnegie Mellon University
23 1.1 oster * Pittsburgh PA 15213-3890
24 1.1 oster *
25 1.1 oster * any improvements or extensions that they make and grant Carnegie the
26 1.1 oster * rights to redistribute these changes.
27 1.1 oster */
28 1.1 oster
29 1.1 oster /**************************************************************************
30 1.1 oster *
31 1.1 oster * map.c -- main code for mapping RAID addresses to physical disk addresses
32 1.1 oster *
33 1.1 oster **************************************************************************/
34 1.9 lukem
35 1.9 lukem #include <sys/cdefs.h>
36 1.31 oster __KERNEL_RCSID(0, "$NetBSD: rf_map.c,v 1.31 2004/03/07 02:59:25 oster Exp $");
37 1.1 oster
38 1.7 oster #include <dev/raidframe/raidframevar.h>
39 1.7 oster
40 1.1 oster #include "rf_threadstuff.h"
41 1.1 oster #include "rf_raid.h"
42 1.1 oster #include "rf_general.h"
43 1.1 oster #include "rf_map.h"
44 1.1 oster #include "rf_shutdown.h"
45 1.1 oster
46 1.27 oster static void rf_FreePDAList(RF_PhysDiskAddr_t *pda_list);
47 1.27 oster static void rf_FreeASMList(RF_AccessStripeMap_t *asm_list);
48 1.1 oster
49 1.21 oster /***************************************************************************
50 1.1 oster *
51 1.21 oster * MapAccess -- main 1st order mapping routine. Maps an access in the
52 1.21 oster * RAID address space to the corresponding set of physical disk
53 1.21 oster * addresses. The result is returned as a list of AccessStripeMap
54 1.21 oster * structures, one per stripe accessed. Each ASM structure contains a
55 1.21 oster * pointer to a list of PhysDiskAddr structures, which describe the
56 1.25 oster * physical locations touched by the user access. Note that this
57 1.25 oster * routine returns only static mapping information, i.e. the list of
58 1.25 oster * physical addresses returned does not necessarily identify the set
59 1.25 oster * of physical locations that will actually be read or written. The
60 1.25 oster * routine also maps the parity. The physical disk location returned
61 1.25 oster * always indicates the entire parity unit, even when only a subset of
62 1.25 oster * it is being accessed. This is because an access that is not stripe
63 1.25 oster * unit aligned but that spans a stripe unit boundary may require
64 1.25 oster * access two distinct portions of the parity unit, and we can't yet
65 1.25 oster * tell which portion(s) we'll actually need. We leave it up to the
66 1.25 oster * algorithm selection code to decide what subset of the parity unit
67 1.25 oster * to access. Note that addresses in the RAID address space must
68 1.25 oster * always be maintained as longs, instead of ints.
69 1.21 oster *
70 1.1 oster * This routine returns NULL if numBlocks is 0
71 1.1 oster *
72 1.26 oster * raidAddress - starting address in RAID address space
73 1.26 oster * numBlocks - number of blocks in RAID address space to access
74 1.26 oster * buffer - buffer to supply/recieve data
75 1.26 oster * remap - 1 => remap address to spare space
76 1.21 oster ***************************************************************************/
77 1.1 oster
78 1.3 oster RF_AccessStripeMapHeader_t *
79 1.26 oster rf_MapAccess(RF_Raid_t *raidPtr, RF_RaidAddr_t raidAddress,
80 1.26 oster RF_SectorCount_t numBlocks, caddr_t buffer, int remap)
81 1.3 oster {
82 1.3 oster RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
83 1.3 oster RF_AccessStripeMapHeader_t *asm_hdr = NULL;
84 1.3 oster RF_AccessStripeMap_t *asm_list = NULL, *asm_p = NULL;
85 1.3 oster int faultsTolerated = layoutPtr->map->faultsTolerated;
86 1.25 oster /* we'll change raidAddress along the way */
87 1.25 oster RF_RaidAddr_t startAddress = raidAddress;
88 1.3 oster RF_RaidAddr_t endAddress = raidAddress + numBlocks;
89 1.23 oster RF_RaidDisk_t *disks = raidPtr->Disks;
90 1.31 oster RF_PhysDiskAddr_t *pda_p;
91 1.31 oster #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
92 1.31 oster RF_PhysDiskAddr_t *pda_q;
93 1.31 oster #endif
94 1.3 oster RF_StripeCount_t numStripes = 0;
95 1.25 oster RF_RaidAddr_t stripeRealEndAddress, stripeEndAddress,
96 1.25 oster nextStripeUnitAddress;
97 1.3 oster RF_RaidAddr_t startAddrWithinStripe, lastRaidAddr;
98 1.3 oster RF_StripeCount_t totStripes;
99 1.3 oster RF_StripeNum_t stripeID, lastSID, SUID, lastSUID;
100 1.3 oster RF_AccessStripeMap_t *asmList, *t_asm;
101 1.3 oster RF_PhysDiskAddr_t *pdaList, *t_pda;
102 1.3 oster
103 1.3 oster /* allocate all the ASMs and PDAs up front */
104 1.3 oster lastRaidAddr = raidAddress + numBlocks - 1;
105 1.3 oster stripeID = rf_RaidAddressToStripeID(layoutPtr, raidAddress);
106 1.3 oster lastSID = rf_RaidAddressToStripeID(layoutPtr, lastRaidAddr);
107 1.3 oster totStripes = lastSID - stripeID + 1;
108 1.3 oster SUID = rf_RaidAddressToStripeUnitID(layoutPtr, raidAddress);
109 1.3 oster lastSUID = rf_RaidAddressToStripeUnitID(layoutPtr, lastRaidAddr);
110 1.3 oster
111 1.3 oster asmList = rf_AllocASMList(totStripes);
112 1.25 oster
113 1.25 oster /* may also need pda(s) per stripe for parity */
114 1.25 oster pdaList = rf_AllocPDAList(lastSUID - SUID + 1 +
115 1.25 oster faultsTolerated * totStripes);
116 1.25 oster
117 1.3 oster
118 1.3 oster if (raidAddress + numBlocks > raidPtr->totalSectors) {
119 1.3 oster RF_ERRORMSG1("Unable to map access because offset (%d) was invalid\n",
120 1.3 oster (int) raidAddress);
121 1.3 oster return (NULL);
122 1.3 oster }
123 1.15 oster #if RF_DEBUG_MAP
124 1.3 oster if (rf_mapDebug)
125 1.3 oster rf_PrintRaidAddressInfo(raidPtr, raidAddress, numBlocks);
126 1.15 oster #endif
127 1.3 oster for (; raidAddress < endAddress;) {
128 1.3 oster /* make the next stripe structure */
129 1.3 oster RF_ASSERT(asmList);
130 1.3 oster t_asm = asmList;
131 1.3 oster asmList = asmList->next;
132 1.6 thorpej memset((char *) t_asm, 0, sizeof(RF_AccessStripeMap_t));
133 1.3 oster if (!asm_p)
134 1.3 oster asm_list = asm_p = t_asm;
135 1.3 oster else {
136 1.3 oster asm_p->next = t_asm;
137 1.3 oster asm_p = asm_p->next;
138 1.3 oster }
139 1.3 oster numStripes++;
140 1.3 oster
141 1.3 oster /* map SUs from current location to the end of the stripe */
142 1.3 oster asm_p->stripeID = /* rf_RaidAddressToStripeID(layoutPtr,
143 1.3 oster raidAddress) */ stripeID++;
144 1.3 oster stripeRealEndAddress = rf_RaidAddressOfNextStripeBoundary(layoutPtr, raidAddress);
145 1.3 oster stripeEndAddress = RF_MIN(endAddress, stripeRealEndAddress);
146 1.3 oster asm_p->raidAddress = raidAddress;
147 1.3 oster asm_p->endRaidAddress = stripeEndAddress;
148 1.3 oster
149 1.3 oster /* map each stripe unit in the stripe */
150 1.3 oster pda_p = NULL;
151 1.25 oster
152 1.25 oster /* Raid addr of start of portion of access that is
153 1.25 oster within this stripe */
154 1.25 oster startAddrWithinStripe = raidAddress;
155 1.25 oster
156 1.3 oster for (; raidAddress < stripeEndAddress;) {
157 1.3 oster RF_ASSERT(pdaList);
158 1.3 oster t_pda = pdaList;
159 1.3 oster pdaList = pdaList->next;
160 1.6 thorpej memset((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
161 1.3 oster if (!pda_p)
162 1.3 oster asm_p->physInfo = pda_p = t_pda;
163 1.3 oster else {
164 1.3 oster pda_p->next = t_pda;
165 1.3 oster pda_p = pda_p->next;
166 1.3 oster }
167 1.3 oster
168 1.3 oster pda_p->type = RF_PDA_TYPE_DATA;
169 1.25 oster (layoutPtr->map->MapSector) (raidPtr, raidAddress,
170 1.25 oster &(pda_p->col),
171 1.25 oster &(pda_p->startSector),
172 1.25 oster remap);
173 1.25 oster
174 1.25 oster /* mark any failures we find. failedPDA is
175 1.25 oster * don't-care if there is more than one
176 1.25 oster * failure */
177 1.25 oster
178 1.25 oster /* the RAID address corresponding to this
179 1.25 oster physical diskaddress */
180 1.25 oster pda_p->raidAddress = raidAddress;
181 1.3 oster nextStripeUnitAddress = rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr, raidAddress);
182 1.3 oster pda_p->numSector = RF_MIN(endAddress, nextStripeUnitAddress) - raidAddress;
183 1.3 oster RF_ASSERT(pda_p->numSector != 0);
184 1.3 oster rf_ASMCheckStatus(raidPtr, pda_p, asm_p, disks, 0);
185 1.3 oster pda_p->bufPtr = buffer + rf_RaidAddressToByte(raidPtr, (raidAddress - startAddress));
186 1.3 oster asm_p->totalSectorsAccessed += pda_p->numSector;
187 1.3 oster asm_p->numStripeUnitsAccessed++;
188 1.3 oster
189 1.3 oster raidAddress = RF_MIN(endAddress, nextStripeUnitAddress);
190 1.3 oster }
191 1.3 oster
192 1.3 oster /* Map the parity. At this stage, the startSector and
193 1.25 oster * numSector fields for the parity unit are always set
194 1.25 oster * to indicate the entire parity unit. We may modify
195 1.25 oster * this after mapping the data portion. */
196 1.3 oster switch (faultsTolerated) {
197 1.3 oster case 0:
198 1.3 oster break;
199 1.3 oster case 1: /* single fault tolerant */
200 1.3 oster RF_ASSERT(pdaList);
201 1.3 oster t_pda = pdaList;
202 1.3 oster pdaList = pdaList->next;
203 1.6 thorpej memset((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
204 1.3 oster pda_p = asm_p->parityInfo = t_pda;
205 1.3 oster pda_p->type = RF_PDA_TYPE_PARITY;
206 1.3 oster (layoutPtr->map->MapParity) (raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
207 1.23 oster &(pda_p->col), &(pda_p->startSector), remap);
208 1.3 oster pda_p->numSector = layoutPtr->sectorsPerStripeUnit;
209 1.3 oster /* raidAddr may be needed to find unit to redirect to */
210 1.3 oster pda_p->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
211 1.3 oster rf_ASMCheckStatus(raidPtr, pda_p, asm_p, disks, 1);
212 1.3 oster rf_ASMParityAdjust(asm_p->parityInfo, startAddrWithinStripe, endAddress, layoutPtr, asm_p);
213 1.3 oster
214 1.3 oster break;
215 1.31 oster #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
216 1.3 oster case 2: /* two fault tolerant */
217 1.3 oster RF_ASSERT(pdaList && pdaList->next);
218 1.3 oster t_pda = pdaList;
219 1.3 oster pdaList = pdaList->next;
220 1.6 thorpej memset((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
221 1.3 oster pda_p = asm_p->parityInfo = t_pda;
222 1.3 oster pda_p->type = RF_PDA_TYPE_PARITY;
223 1.3 oster t_pda = pdaList;
224 1.3 oster pdaList = pdaList->next;
225 1.6 thorpej memset((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
226 1.3 oster pda_q = asm_p->qInfo = t_pda;
227 1.3 oster pda_q->type = RF_PDA_TYPE_Q;
228 1.3 oster (layoutPtr->map->MapParity) (raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
229 1.23 oster &(pda_p->col), &(pda_p->startSector), remap);
230 1.3 oster (layoutPtr->map->MapQ) (raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
231 1.23 oster &(pda_q->col), &(pda_q->startSector), remap);
232 1.3 oster pda_q->numSector = pda_p->numSector = layoutPtr->sectorsPerStripeUnit;
233 1.3 oster /* raidAddr may be needed to find unit to redirect to */
234 1.3 oster pda_p->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
235 1.3 oster pda_q->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
236 1.3 oster /* failure mode stuff */
237 1.3 oster rf_ASMCheckStatus(raidPtr, pda_p, asm_p, disks, 1);
238 1.3 oster rf_ASMCheckStatus(raidPtr, pda_q, asm_p, disks, 1);
239 1.3 oster rf_ASMParityAdjust(asm_p->parityInfo, startAddrWithinStripe, endAddress, layoutPtr, asm_p);
240 1.3 oster rf_ASMParityAdjust(asm_p->qInfo, startAddrWithinStripe, endAddress, layoutPtr, asm_p);
241 1.3 oster break;
242 1.31 oster #endif
243 1.3 oster }
244 1.3 oster }
245 1.3 oster RF_ASSERT(asmList == NULL && pdaList == NULL);
246 1.3 oster /* make the header structure */
247 1.3 oster asm_hdr = rf_AllocAccessStripeMapHeader();
248 1.3 oster RF_ASSERT(numStripes == totStripes);
249 1.3 oster asm_hdr->numStripes = numStripes;
250 1.3 oster asm_hdr->stripeMap = asm_list;
251 1.3 oster
252 1.15 oster #if RF_DEBUG_MAP
253 1.3 oster if (rf_mapDebug)
254 1.3 oster rf_PrintAccessStripeMap(asm_hdr);
255 1.15 oster #endif
256 1.3 oster return (asm_hdr);
257 1.1 oster }
258 1.21 oster
259 1.21 oster /***************************************************************************
260 1.21 oster * This routine walks through an ASM list and marks the PDAs that have
261 1.21 oster * failed. It's called only when a disk failure causes an in-flight
262 1.21 oster * DAG to fail. The parity may consist of two components, but we want
263 1.21 oster * to use only one failedPDA pointer. Thus we set failedPDA to point
264 1.21 oster * to the first parity component, and rely on the rest of the code to
265 1.21 oster * do the right thing with this.
266 1.21 oster ***************************************************************************/
267 1.1 oster
268 1.3 oster void
269 1.26 oster rf_MarkFailuresInASMList(RF_Raid_t *raidPtr,
270 1.26 oster RF_AccessStripeMapHeader_t *asm_h)
271 1.3 oster {
272 1.23 oster RF_RaidDisk_t *disks = raidPtr->Disks;
273 1.3 oster RF_AccessStripeMap_t *asmap;
274 1.3 oster RF_PhysDiskAddr_t *pda;
275 1.3 oster
276 1.3 oster for (asmap = asm_h->stripeMap; asmap; asmap = asmap->next) {
277 1.25 oster asmap->numDataFailed = 0;
278 1.25 oster asmap->numParityFailed = 0;
279 1.25 oster asmap->numQFailed = 0;
280 1.3 oster asmap->numFailedPDAs = 0;
281 1.6 thorpej memset((char *) asmap->failedPDAs, 0,
282 1.3 oster RF_MAX_FAILED_PDA * sizeof(RF_PhysDiskAddr_t *));
283 1.3 oster for (pda = asmap->physInfo; pda; pda = pda->next) {
284 1.23 oster if (RF_DEAD_DISK(disks[pda->col].status)) {
285 1.3 oster asmap->numDataFailed++;
286 1.3 oster asmap->failedPDAs[asmap->numFailedPDAs] = pda;
287 1.3 oster asmap->numFailedPDAs++;
288 1.3 oster }
289 1.3 oster }
290 1.3 oster pda = asmap->parityInfo;
291 1.23 oster if (pda && RF_DEAD_DISK(disks[pda->col].status)) {
292 1.3 oster asmap->numParityFailed++;
293 1.3 oster asmap->failedPDAs[asmap->numFailedPDAs] = pda;
294 1.3 oster asmap->numFailedPDAs++;
295 1.3 oster }
296 1.3 oster pda = asmap->qInfo;
297 1.23 oster if (pda && RF_DEAD_DISK(disks[pda->col].status)) {
298 1.3 oster asmap->numQFailed++;
299 1.3 oster asmap->failedPDAs[asmap->numFailedPDAs] = pda;
300 1.3 oster asmap->numFailedPDAs++;
301 1.3 oster }
302 1.3 oster }
303 1.1 oster }
304 1.3 oster
305 1.21 oster /***************************************************************************
306 1.1 oster *
307 1.21 oster * routines to allocate and free list elements. All allocation
308 1.21 oster * routines zero the structure before returning it.
309 1.1 oster *
310 1.21 oster * FreePhysDiskAddr is static. It should never be called directly,
311 1.21 oster * because FreeAccessStripeMap takes care of freeing the PhysDiskAddr
312 1.21 oster * list.
313 1.1 oster *
314 1.21 oster ***************************************************************************/
315 1.1 oster
316 1.24 oster static struct pool rf_asmhdr_pool;
317 1.1 oster #define RF_MAX_FREE_ASMHDR 128
318 1.30 oster #define RF_MIN_FREE_ASMHDR 32
319 1.1 oster
320 1.24 oster static struct pool rf_asm_pool;
321 1.1 oster #define RF_MAX_FREE_ASM 192
322 1.30 oster #define RF_MIN_FREE_ASM 64
323 1.1 oster
324 1.24 oster static struct pool rf_pda_pool; /* may need to be visible for
325 1.24 oster rf_dagdegrd.c and rf_dagdegwr.c,
326 1.24 oster if they can be convinced to free
327 1.24 oster the space easily */
328 1.1 oster #define RF_MAX_FREE_PDA 192
329 1.30 oster #define RF_MIN_FREE_PDA 64
330 1.1 oster
331 1.25 oster /* called at shutdown time. So far, all that is necessary is to
332 1.25 oster release all the free lists */
333 1.1 oster static void rf_ShutdownMapModule(void *);
334 1.3 oster static void
335 1.26 oster rf_ShutdownMapModule(void *ignored)
336 1.1 oster {
337 1.24 oster pool_destroy(&rf_asmhdr_pool);
338 1.24 oster pool_destroy(&rf_asm_pool);
339 1.24 oster pool_destroy(&rf_pda_pool);
340 1.1 oster }
341 1.1 oster
342 1.3 oster int
343 1.26 oster rf_ConfigureMapModule(RF_ShutdownList_t **listp)
344 1.1 oster {
345 1.1 oster
346 1.24 oster pool_init(&rf_asmhdr_pool, sizeof(RF_AccessStripeMapHeader_t),
347 1.24 oster 0, 0, 0, "rf_asmhdr_pl", NULL);
348 1.24 oster pool_sethiwat(&rf_asmhdr_pool, RF_MAX_FREE_ASMHDR);
349 1.30 oster pool_prime(&rf_asmhdr_pool, RF_MIN_FREE_ASMHDR);
350 1.30 oster pool_setlowat(&rf_asmhdr_pool, RF_MIN_FREE_ASMHDR);
351 1.24 oster
352 1.24 oster pool_init(&rf_asm_pool, sizeof(RF_AccessStripeMap_t),
353 1.24 oster 0, 0, 0, "rf_asm_pl", NULL);
354 1.24 oster pool_sethiwat(&rf_asm_pool, RF_MAX_FREE_ASM);
355 1.30 oster pool_prime(&rf_asm_pool, RF_MIN_FREE_ASM);
356 1.30 oster pool_setlowat(&rf_asm_pool, RF_MIN_FREE_ASM);
357 1.24 oster
358 1.24 oster pool_init(&rf_pda_pool, sizeof(RF_PhysDiskAddr_t),
359 1.24 oster 0, 0, 0, "rf_pda_pl", NULL);
360 1.24 oster pool_sethiwat(&rf_pda_pool, RF_MAX_FREE_PDA);
361 1.30 oster pool_prime(&rf_pda_pool, RF_MIN_FREE_PDA);
362 1.30 oster pool_setlowat(&rf_pda_pool, RF_MIN_FREE_PDA);
363 1.24 oster
364 1.29 oster rf_ShutdownCreate(listp, rf_ShutdownMapModule, NULL);
365 1.29 oster
366 1.3 oster return (0);
367 1.1 oster }
368 1.1 oster
369 1.3 oster RF_AccessStripeMapHeader_t *
370 1.3 oster rf_AllocAccessStripeMapHeader()
371 1.1 oster {
372 1.1 oster RF_AccessStripeMapHeader_t *p;
373 1.1 oster
374 1.24 oster p = pool_get(&rf_asmhdr_pool, PR_WAITOK);
375 1.6 thorpej memset((char *) p, 0, sizeof(RF_AccessStripeMapHeader_t));
376 1.1 oster
377 1.3 oster return (p);
378 1.1 oster }
379 1.1 oster
380 1.3 oster void
381 1.26 oster rf_FreeAccessStripeMapHeader(RF_AccessStripeMapHeader_t *p)
382 1.1 oster {
383 1.24 oster pool_put(&rf_asmhdr_pool, p);
384 1.1 oster }
385 1.1 oster
386 1.3 oster RF_PhysDiskAddr_t *
387 1.3 oster rf_AllocPhysDiskAddr()
388 1.1 oster {
389 1.1 oster RF_PhysDiskAddr_t *p;
390 1.1 oster
391 1.24 oster p = pool_get(&rf_pda_pool, PR_WAITOK);
392 1.6 thorpej memset((char *) p, 0, sizeof(RF_PhysDiskAddr_t));
393 1.1 oster
394 1.3 oster return (p);
395 1.1 oster }
396 1.25 oster /* allocates a list of PDAs, locking the free list only once when we
397 1.25 oster * have to call calloc, we do it one component at a time to simplify
398 1.25 oster * the process of freeing the list at program shutdown. This should
399 1.25 oster * not be much of a performance hit, because it should be very
400 1.25 oster * infrequently executed. */
401 1.3 oster RF_PhysDiskAddr_t *
402 1.26 oster rf_AllocPDAList(int count)
403 1.1 oster {
404 1.24 oster RF_PhysDiskAddr_t *p, *prev;
405 1.24 oster int i;
406 1.24 oster
407 1.24 oster p = NULL;
408 1.24 oster prev = NULL;
409 1.24 oster for (i = 0; i < count; i++) {
410 1.24 oster p = pool_get(&rf_pda_pool, PR_WAITOK);
411 1.24 oster p->next = prev;
412 1.24 oster prev = p;
413 1.24 oster }
414 1.1 oster
415 1.3 oster return (p);
416 1.1 oster }
417 1.1 oster
418 1.18 oster #if RF_INCLUDE_PARITYLOGGING > 0
419 1.3 oster void
420 1.26 oster rf_FreePhysDiskAddr(RF_PhysDiskAddr_t *p)
421 1.1 oster {
422 1.24 oster pool_put(&rf_pda_pool, p);
423 1.1 oster }
424 1.18 oster #endif
425 1.1 oster
426 1.3 oster static void
427 1.27 oster rf_FreePDAList(RF_PhysDiskAddr_t *pda_list)
428 1.1 oster {
429 1.24 oster RF_PhysDiskAddr_t *p, *tmp;
430 1.24 oster
431 1.27 oster p=pda_list;
432 1.24 oster while (p) {
433 1.24 oster tmp = p->next;
434 1.24 oster pool_put(&rf_pda_pool, p);
435 1.24 oster p = tmp;
436 1.24 oster }
437 1.1 oster }
438 1.1 oster
439 1.25 oster /* this is essentially identical to AllocPDAList. I should combine
440 1.25 oster * the two. when we have to call calloc, we do it one component at a
441 1.25 oster * time to simplify the process of freeing the list at program
442 1.25 oster * shutdown. This should not be much of a performance hit, because it
443 1.25 oster * should be very infrequently executed. */
444 1.3 oster RF_AccessStripeMap_t *
445 1.26 oster rf_AllocASMList(int count)
446 1.1 oster {
447 1.24 oster RF_AccessStripeMap_t *p, *prev;
448 1.24 oster int i;
449 1.1 oster
450 1.24 oster p = NULL;
451 1.24 oster prev = NULL;
452 1.24 oster for (i = 0; i < count; i++) {
453 1.24 oster p = pool_get(&rf_asm_pool, PR_WAITOK);
454 1.24 oster p->next = prev;
455 1.24 oster prev = p;
456 1.24 oster }
457 1.3 oster return (p);
458 1.1 oster }
459 1.1 oster
460 1.3 oster static void
461 1.27 oster rf_FreeASMList(RF_AccessStripeMap_t *asm_list)
462 1.3 oster {
463 1.24 oster RF_AccessStripeMap_t *p, *tmp;
464 1.24 oster
465 1.27 oster p=asm_list;
466 1.24 oster while (p) {
467 1.24 oster tmp = p->next;
468 1.24 oster pool_put(&rf_asm_pool, p);
469 1.24 oster p = tmp;
470 1.24 oster }
471 1.3 oster }
472 1.3 oster
473 1.3 oster void
474 1.26 oster rf_FreeAccessStripeMap(RF_AccessStripeMapHeader_t *hdr)
475 1.3 oster {
476 1.28 oster RF_AccessStripeMap_t *p;
477 1.3 oster RF_PhysDiskAddr_t *pdp, *trailer, *pdaList = NULL, *pdaEnd = NULL;
478 1.3 oster int count = 0, t, asm_count = 0;
479 1.3 oster
480 1.3 oster for (p = hdr->stripeMap; p; p = p->next) {
481 1.3 oster
482 1.3 oster /* link the 3 pda lists into the accumulating pda list */
483 1.3 oster
484 1.3 oster if (!pdaList)
485 1.3 oster pdaList = p->qInfo;
486 1.3 oster else
487 1.3 oster pdaEnd->next = p->qInfo;
488 1.3 oster for (trailer = NULL, pdp = p->qInfo; pdp;) {
489 1.3 oster trailer = pdp;
490 1.3 oster pdp = pdp->next;
491 1.3 oster count++;
492 1.3 oster }
493 1.3 oster if (trailer)
494 1.3 oster pdaEnd = trailer;
495 1.3 oster
496 1.3 oster if (!pdaList)
497 1.3 oster pdaList = p->parityInfo;
498 1.3 oster else
499 1.3 oster pdaEnd->next = p->parityInfo;
500 1.3 oster for (trailer = NULL, pdp = p->parityInfo; pdp;) {
501 1.3 oster trailer = pdp;
502 1.3 oster pdp = pdp->next;
503 1.3 oster count++;
504 1.3 oster }
505 1.3 oster if (trailer)
506 1.3 oster pdaEnd = trailer;
507 1.3 oster
508 1.3 oster if (!pdaList)
509 1.3 oster pdaList = p->physInfo;
510 1.3 oster else
511 1.3 oster pdaEnd->next = p->physInfo;
512 1.3 oster for (trailer = NULL, pdp = p->physInfo; pdp;) {
513 1.3 oster trailer = pdp;
514 1.3 oster pdp = pdp->next;
515 1.3 oster count++;
516 1.3 oster }
517 1.3 oster if (trailer)
518 1.3 oster pdaEnd = trailer;
519 1.3 oster
520 1.3 oster asm_count++;
521 1.3 oster }
522 1.3 oster
523 1.3 oster /* debug only */
524 1.3 oster for (t = 0, pdp = pdaList; pdp; pdp = pdp->next)
525 1.3 oster t++;
526 1.3 oster RF_ASSERT(t == count);
527 1.3 oster
528 1.3 oster if (pdaList)
529 1.27 oster rf_FreePDAList(pdaList);
530 1.27 oster rf_FreeASMList(hdr->stripeMap);
531 1.3 oster rf_FreeAccessStripeMapHeader(hdr);
532 1.1 oster }
533 1.21 oster /* We can't use the large write optimization if there are any failures
534 1.21 oster * in the stripe. In the declustered layout, there is no way to
535 1.21 oster * immediately determine what disks constitute a stripe, so we
536 1.21 oster * actually have to hunt through the stripe looking for failures. The
537 1.21 oster * reason we map the parity instead of just using asm->parityInfo->col
538 1.21 oster * is because the latter may have been already redirected to a spare
539 1.21 oster * drive, which would mess up the computation of the stripe offset.
540 1.1 oster *
541 1.21 oster * ASSUMES AT MOST ONE FAILURE IN THE STRIPE. */
542 1.3 oster int
543 1.26 oster rf_CheckStripeForFailures(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap)
544 1.3 oster {
545 1.23 oster RF_RowCol_t tcol, pcol, *diskids, i;
546 1.3 oster RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
547 1.3 oster RF_StripeCount_t stripeOffset;
548 1.3 oster int numFailures;
549 1.3 oster RF_RaidAddr_t sosAddr;
550 1.3 oster RF_SectorNum_t diskOffset, poffset;
551 1.3 oster
552 1.3 oster /* quick out in the fault-free case. */
553 1.3 oster RF_LOCK_MUTEX(raidPtr->mutex);
554 1.3 oster numFailures = raidPtr->numFailures;
555 1.3 oster RF_UNLOCK_MUTEX(raidPtr->mutex);
556 1.3 oster if (numFailures == 0)
557 1.3 oster return (0);
558 1.3 oster
559 1.25 oster sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr,
560 1.25 oster asmap->raidAddress);
561 1.25 oster (layoutPtr->map->IdentifyStripe) (raidPtr, asmap->raidAddress,
562 1.25 oster &diskids);
563 1.25 oster (layoutPtr->map->MapParity) (raidPtr, asmap->raidAddress,
564 1.25 oster &pcol, &poffset, 0); /* get pcol */
565 1.3 oster
566 1.25 oster /* this need not be true if we've redirected the access to a
567 1.25 oster * spare in another row RF_ASSERT(row == testrow); */
568 1.3 oster stripeOffset = 0;
569 1.3 oster for (i = 0; i < layoutPtr->numDataCol + layoutPtr->numParityCol; i++) {
570 1.3 oster if (diskids[i] != pcol) {
571 1.23 oster if (RF_DEAD_DISK(raidPtr->Disks[diskids[i]].status)) {
572 1.23 oster if (raidPtr->status != rf_rs_reconstructing)
573 1.3 oster return (1);
574 1.23 oster RF_ASSERT(raidPtr->reconControl->fcol == diskids[i]);
575 1.3 oster layoutPtr->map->MapSector(raidPtr,
576 1.3 oster sosAddr + stripeOffset * layoutPtr->sectorsPerStripeUnit,
577 1.23 oster &tcol, &diskOffset, 0);
578 1.23 oster RF_ASSERT(tcol == diskids[i]);
579 1.23 oster if (!rf_CheckRUReconstructed(raidPtr->reconControl->reconMap, diskOffset))
580 1.3 oster return (1);
581 1.3 oster asmap->flags |= RF_ASM_REDIR_LARGE_WRITE;
582 1.3 oster return (0);
583 1.3 oster }
584 1.3 oster stripeOffset++;
585 1.3 oster }
586 1.3 oster }
587 1.3 oster return (0);
588 1.1 oster }
589 1.18 oster #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) || (RF_INCLUDE_EVENODD >0)
590 1.1 oster /*
591 1.1 oster return the number of failed data units in the stripe.
592 1.1 oster */
593 1.1 oster
594 1.3 oster int
595 1.26 oster rf_NumFailedDataUnitsInStripe(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap)
596 1.3 oster {
597 1.3 oster RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
598 1.23 oster RF_RowCol_t tcol, i;
599 1.3 oster RF_SectorNum_t diskOffset;
600 1.3 oster RF_RaidAddr_t sosAddr;
601 1.3 oster int numFailures;
602 1.3 oster
603 1.3 oster /* quick out in the fault-free case. */
604 1.3 oster RF_LOCK_MUTEX(raidPtr->mutex);
605 1.3 oster numFailures = raidPtr->numFailures;
606 1.3 oster RF_UNLOCK_MUTEX(raidPtr->mutex);
607 1.3 oster if (numFailures == 0)
608 1.3 oster return (0);
609 1.3 oster numFailures = 0;
610 1.3 oster
611 1.25 oster sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr,
612 1.25 oster asmap->raidAddress);
613 1.3 oster for (i = 0; i < layoutPtr->numDataCol; i++) {
614 1.3 oster (layoutPtr->map->MapSector) (raidPtr, sosAddr + i * layoutPtr->sectorsPerStripeUnit,
615 1.3 oster &trow, &tcol, &diskOffset, 0);
616 1.23 oster if (RF_DEAD_DISK(raidPtr->Disks[tcol].status))
617 1.3 oster numFailures++;
618 1.3 oster }
619 1.1 oster
620 1.3 oster return numFailures;
621 1.1 oster }
622 1.18 oster #endif
623 1.1 oster
624 1.25 oster /****************************************************************************
625 1.1 oster *
626 1.1 oster * debug routines
627 1.1 oster *
628 1.25 oster ***************************************************************************/
629 1.18 oster #if RF_DEBUG_MAP
630 1.3 oster void
631 1.26 oster rf_PrintAccessStripeMap(RF_AccessStripeMapHeader_t *asm_h)
632 1.1 oster {
633 1.3 oster rf_PrintFullAccessStripeMap(asm_h, 0);
634 1.1 oster }
635 1.18 oster #endif
636 1.1 oster
637 1.26 oster /* prbuf - flag to print buffer pointers */
638 1.3 oster void
639 1.26 oster rf_PrintFullAccessStripeMap(RF_AccessStripeMapHeader_t *asm_h, int prbuf)
640 1.3 oster {
641 1.3 oster int i;
642 1.3 oster RF_AccessStripeMap_t *asmap = asm_h->stripeMap;
643 1.3 oster RF_PhysDiskAddr_t *p;
644 1.3 oster printf("%d stripes total\n", (int) asm_h->numStripes);
645 1.3 oster for (; asmap; asmap = asmap->next) {
646 1.3 oster /* printf("Num failures: %d\n",asmap->numDataFailed); */
647 1.3 oster /* printf("Num sectors:
648 1.3 oster * %d\n",(int)asmap->totalSectorsAccessed); */
649 1.3 oster printf("Stripe %d (%d sectors), failures: %d data, %d parity: ",
650 1.3 oster (int) asmap->stripeID,
651 1.3 oster (int) asmap->totalSectorsAccessed,
652 1.3 oster (int) asmap->numDataFailed,
653 1.3 oster (int) asmap->numParityFailed);
654 1.3 oster if (asmap->parityInfo) {
655 1.23 oster printf("Parity [c%d s%d-%d", asmap->parityInfo->col,
656 1.3 oster (int) asmap->parityInfo->startSector,
657 1.3 oster (int) (asmap->parityInfo->startSector +
658 1.3 oster asmap->parityInfo->numSector - 1));
659 1.3 oster if (prbuf)
660 1.3 oster printf(" b0x%lx", (unsigned long) asmap->parityInfo->bufPtr);
661 1.3 oster if (asmap->parityInfo->next) {
662 1.23 oster printf(", c%d s%d-%d", asmap->parityInfo->next->col,
663 1.3 oster (int) asmap->parityInfo->next->startSector,
664 1.3 oster (int) (asmap->parityInfo->next->startSector +
665 1.3 oster asmap->parityInfo->next->numSector - 1));
666 1.3 oster if (prbuf)
667 1.3 oster printf(" b0x%lx", (unsigned long) asmap->parityInfo->next->bufPtr);
668 1.3 oster RF_ASSERT(asmap->parityInfo->next->next == NULL);
669 1.3 oster }
670 1.3 oster printf("]\n\t");
671 1.3 oster }
672 1.3 oster for (i = 0, p = asmap->physInfo; p; p = p->next, i++) {
673 1.23 oster printf("SU c%d s%d-%d ", p->col, (int) p->startSector,
674 1.3 oster (int) (p->startSector + p->numSector - 1));
675 1.3 oster if (prbuf)
676 1.3 oster printf("b0x%lx ", (unsigned long) p->bufPtr);
677 1.3 oster if (i && !(i & 1))
678 1.3 oster printf("\n\t");
679 1.3 oster }
680 1.3 oster printf("\n");
681 1.3 oster p = asm_h->stripeMap->failedPDAs[0];
682 1.3 oster if (asm_h->stripeMap->numDataFailed + asm_h->stripeMap->numParityFailed > 1)
683 1.3 oster printf("[multiple failures]\n");
684 1.3 oster else
685 1.3 oster if (asm_h->stripeMap->numDataFailed + asm_h->stripeMap->numParityFailed > 0)
686 1.23 oster printf("\t[Failed PDA: c%d s%d-%d]\n", p->col,
687 1.3 oster (int) p->startSector, (int) (p->startSector + p->numSector - 1));
688 1.3 oster }
689 1.1 oster }
690 1.1 oster
691 1.15 oster #if RF_MAP_DEBUG
692 1.3 oster void
693 1.26 oster rf_PrintRaidAddressInfo(RF_Raid_t *raidPtr, RF_RaidAddr_t raidAddr,
694 1.26 oster RF_SectorCount_t numBlocks)
695 1.3 oster {
696 1.3 oster RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
697 1.3 oster RF_RaidAddr_t ra, sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, raidAddr);
698 1.3 oster
699 1.3 oster printf("Raid addrs of SU boundaries from start of stripe to end of access:\n\t");
700 1.3 oster for (ra = sosAddr; ra <= raidAddr + numBlocks; ra += layoutPtr->sectorsPerStripeUnit) {
701 1.3 oster printf("%d (0x%x), ", (int) ra, (int) ra);
702 1.3 oster }
703 1.3 oster printf("\n");
704 1.3 oster printf("Offset into stripe unit: %d (0x%x)\n",
705 1.3 oster (int) (raidAddr % layoutPtr->sectorsPerStripeUnit),
706 1.3 oster (int) (raidAddr % layoutPtr->sectorsPerStripeUnit));
707 1.3 oster }
708 1.15 oster #endif
709 1.25 oster /* given a parity descriptor and the starting address within a stripe,
710 1.25 oster * range restrict the parity descriptor to touch only the correct
711 1.25 oster * stuff. */
712 1.3 oster void
713 1.26 oster rf_ASMParityAdjust(RF_PhysDiskAddr_t *toAdjust,
714 1.26 oster RF_StripeNum_t startAddrWithinStripe,
715 1.26 oster RF_SectorNum_t endAddress,
716 1.26 oster RF_RaidLayout_t *layoutPtr,
717 1.26 oster RF_AccessStripeMap_t *asm_p)
718 1.3 oster {
719 1.3 oster RF_PhysDiskAddr_t *new_pda;
720 1.3 oster
721 1.25 oster /* when we're accessing only a portion of one stripe unit, we
722 1.25 oster * want the parity descriptor to identify only the chunk of
723 1.25 oster * parity associated with the data. When the access spans
724 1.25 oster * exactly one stripe unit boundary and is less than a stripe
725 1.25 oster * unit in size, it uses two disjoint regions of the parity
726 1.25 oster * unit. When an access spans more than one stripe unit
727 1.25 oster * boundary, it uses all of the parity unit.
728 1.3 oster *
729 1.25 oster * To better handle the case where stripe units are small, we
730 1.25 oster * may eventually want to change the 2nd case so that if the
731 1.25 oster * SU size is below some threshold, we just read/write the
732 1.25 oster * whole thing instead of breaking it up into two accesses. */
733 1.3 oster if (asm_p->numStripeUnitsAccessed == 1) {
734 1.3 oster int x = (startAddrWithinStripe % layoutPtr->sectorsPerStripeUnit);
735 1.3 oster toAdjust->startSector += x;
736 1.3 oster toAdjust->raidAddress += x;
737 1.3 oster toAdjust->numSector = asm_p->physInfo->numSector;
738 1.3 oster RF_ASSERT(toAdjust->numSector != 0);
739 1.3 oster } else
740 1.3 oster if (asm_p->numStripeUnitsAccessed == 2 && asm_p->totalSectorsAccessed < layoutPtr->sectorsPerStripeUnit) {
741 1.3 oster int x = (startAddrWithinStripe % layoutPtr->sectorsPerStripeUnit);
742 1.3 oster
743 1.3 oster /* create a second pda and copy the parity map info
744 1.3 oster * into it */
745 1.3 oster RF_ASSERT(toAdjust->next == NULL);
746 1.3 oster new_pda = toAdjust->next = rf_AllocPhysDiskAddr();
747 1.3 oster *new_pda = *toAdjust; /* structure assignment */
748 1.3 oster new_pda->next = NULL;
749 1.3 oster
750 1.3 oster /* adjust the start sector & number of blocks for the
751 1.3 oster * first parity pda */
752 1.3 oster toAdjust->startSector += x;
753 1.3 oster toAdjust->raidAddress += x;
754 1.3 oster toAdjust->numSector = rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr, startAddrWithinStripe) - startAddrWithinStripe;
755 1.3 oster RF_ASSERT(toAdjust->numSector != 0);
756 1.3 oster
757 1.3 oster /* adjust the second pda */
758 1.3 oster new_pda->numSector = endAddress - rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, endAddress);
759 1.3 oster /* new_pda->raidAddress =
760 1.3 oster * rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr,
761 1.3 oster * toAdjust->raidAddress); */
762 1.3 oster RF_ASSERT(new_pda->numSector != 0);
763 1.3 oster }
764 1.1 oster }
765 1.11 oster
766 1.25 oster /* Check if a disk has been spared or failed. If spared, redirect the
767 1.25 oster * I/O. If it has been failed, record it in the asm pointer. Fourth
768 1.25 oster * arg is whether data or parity. */
769 1.3 oster void
770 1.26 oster rf_ASMCheckStatus(RF_Raid_t *raidPtr, RF_PhysDiskAddr_t *pda_p,
771 1.26 oster RF_AccessStripeMap_t *asm_p, RF_RaidDisk_t *disks,
772 1.26 oster int parity)
773 1.3 oster {
774 1.3 oster RF_DiskStatus_t dstatus;
775 1.23 oster RF_RowCol_t fcol;
776 1.3 oster
777 1.23 oster dstatus = disks[pda_p->col].status;
778 1.3 oster
779 1.3 oster if (dstatus == rf_ds_spared) {
780 1.3 oster /* if the disk has been spared, redirect access to the spare */
781 1.3 oster fcol = pda_p->col;
782 1.23 oster pda_p->col = disks[fcol].spareCol;
783 1.3 oster } else
784 1.3 oster if (dstatus == rf_ds_dist_spared) {
785 1.3 oster /* ditto if disk has been spared to dist spare space */
786 1.15 oster #if RF_DEBUG_MAP
787 1.23 oster RF_RowCol_t oc = pda_p->col;
788 1.3 oster RF_SectorNum_t oo = pda_p->startSector;
789 1.15 oster #endif
790 1.3 oster if (pda_p->type == RF_PDA_TYPE_DATA)
791 1.23 oster raidPtr->Layout.map->MapSector(raidPtr, pda_p->raidAddress, &pda_p->col, &pda_p->startSector, RF_REMAP);
792 1.3 oster else
793 1.23 oster raidPtr->Layout.map->MapParity(raidPtr, pda_p->raidAddress, &pda_p->col, &pda_p->startSector, RF_REMAP);
794 1.3 oster
795 1.15 oster #if RF_DEBUG_MAP
796 1.3 oster if (rf_mapDebug) {
797 1.23 oster printf("Redirected c %d o %d -> c %d o %d\n", oc, (int) oo,
798 1.23 oster pda_p->col, (int) pda_p->startSector);
799 1.3 oster }
800 1.15 oster #endif
801 1.3 oster } else
802 1.3 oster if (RF_DEAD_DISK(dstatus)) {
803 1.3 oster /* if the disk is inaccessible, mark the
804 1.3 oster * failure */
805 1.3 oster if (parity)
806 1.3 oster asm_p->numParityFailed++;
807 1.3 oster else {
808 1.3 oster asm_p->numDataFailed++;
809 1.3 oster }
810 1.3 oster asm_p->failedPDAs[asm_p->numFailedPDAs] = pda_p;
811 1.3 oster asm_p->numFailedPDAs++;
812 1.1 oster #if 0
813 1.3 oster switch (asm_p->numParityFailed + asm_p->numDataFailed) {
814 1.3 oster case 1:
815 1.3 oster asm_p->failedPDAs[0] = pda_p;
816 1.3 oster break;
817 1.3 oster case 2:
818 1.3 oster asm_p->failedPDAs[1] = pda_p;
819 1.3 oster default:
820 1.3 oster break;
821 1.3 oster }
822 1.1 oster #endif
823 1.3 oster }
824 1.3 oster /* the redirected access should never span a stripe unit boundary */
825 1.3 oster RF_ASSERT(rf_RaidAddressToStripeUnitID(&raidPtr->Layout, pda_p->raidAddress) ==
826 1.3 oster rf_RaidAddressToStripeUnitID(&raidPtr->Layout, pda_p->raidAddress + pda_p->numSector - 1));
827 1.3 oster RF_ASSERT(pda_p->col != -1);
828 1.1 oster }
829